CN114144688A

CN114144688A - System and method for sensing deformation of magnetic material and method of making the same

Info

Publication number: CN114144688A
Application number: CN202080052939.6A
Authority: CN
Inventors: 泰丝·赫勒布瑞科斯; 卡梅尔·马吉德
Original assignee: Carnegie Mellon University
Current assignee: Carnegie Mellon University
Priority date: 2019-06-21
Filing date: 2020-06-22
Publication date: 2022-03-04
Also published as: EP3973307A4; JP2022541997A; EP3973307A2; KR20220024447A; WO2021061240A2; WO2021061240A3; US20220349695A1

Abstract

A soft magnetic sensor comprising a soft material containing randomly distributed magnetic particles and a magnetometer capable of estimating force and localizing contact in a continuous area. The reference magnetometer may be used to filter motion noise and ambient noise. The method of locating contact and determining force includes data analysis of the output of the magnetometer. In some embodiments, the sensor may position the object prior to contact.

Description

System and method for sensing deformation of magnetic material and method of making the same

Cross reference to related applications

According to 35 u.s.c. § 119, the present application claims the benefit of provisional application No. 62/864,766 filed 2019 on 6, 21, which is incorporated herein by reference.

Declaration of federally sponsored research

The invention was made with government support in accordance with N00014-16-1-2301, issued by the naval research institute. The government has certain rights in this invention.

Background

The technology of the present disclosure relates generally to sensing. More particularly, the present invention relates to soft measurements that use deformation of magnetic materials to provide feedback about the environment in which they are located.

The continuing development of wearable technology, soft robotics and human-computer interaction has led to an increased interest in sensors. The inability to accurately determine the position of an object using these techniques can affect the ability to perform certain functions. For example, for robotic systems, inaccuracies can prevent the robot from positioning and operating a tool or other object. Vision-based sensing is good at finding objects in the work area, but does not provide guidance within 1 to 2 centimeters from the target object. Additionally, vision-based systems may also perform poorly if the camera is blocked or the surface is reflective or transparent.

To overcome the limitations of vision-based systems, tactile sensors provide critical information about the environment surrounding the tactile sensor by measuring the contact force. Since the tactile sensor is based on touch, it provides information only after contacting an object and cannot assist in approaching a target object. Soft tactile sensors are a sub-class of tactile sensors that employ deformable and compliant materials at the interaction surface. Soft sensors provide not only rich environmental information, but also effective mechanical properties to achieve successful robotic manipulation, human-computer interaction, and material classification. Soft touch sensors may use a variety of switching modes, such as optical, resistive, and capacitive. Although soft tactile sensors may provide greater accuracy than vision-based sensors, their wide implementation is limited due to inextensible manufacturing techniques, lack of personalized customization functionality, and complex integration requirements. For example, when using resistive or capacitive soft sensors, the increase in density of each cell is associated with unmanageable scaling and weak soft-hard electrical interface failures in the wiring. In addition, as with other tactile sensors, soft tactile sensors provide information only immediately upon contact with an object.

Magnetic induction can overcome several of these obstacles because magnetic induction has a limited dependence on direct wire, which provides high resolution, high speed sensing by measuring changes in magnetic flux or electromagnetic induction. Additionally, in some applications, magnetic induction may provide a sensor output prior to contact. Despite these improvements over other types of sensors, magnetic sensors are susceptible to ambient magnetic noise. Furthermore, when implemented as a soft sensor, material failure may occur at the junction between the rigid magnet and the soft elastomer used for the sensor, limiting the technology to non-soft sensing applications. For example, a common magnetic sensor combines a hall effect sensing chip with a discrete permanent magnet suspended between two elastomer layers. It would therefore be advantageous to develop a sensing system that overcomes these limitations to provide a tactile surface for single point contact positioning, and provides fast positioning and force estimation in free space.

Disclosure of Invention

Aspects disclosed in the detailed description include a soft sensor, a method for sensing deformation of a magnetic material, and a method of manufacturing the same. Related methods and systems are also disclosed.

In at least one non-limiting embodiment, a soft magnetic sensor is provided that includes a soft material containing randomly distributed magnetic particles and a magnetometer capable of estimating force and localized contact over a continuous area. In one example, the sensor coversAbout 15 to 40mm²A continuous region of (a). In some embodiments discussed herein, the force and localized contact are estimated using an integrated circuit that performs data analysis on the output from the magnetometer. In some embodiments, the magnetic material or "skin" is composed of a silicone elastomer loaded with magnetic particles. When the elastomer deforms, a portion of the embedded magnetic particles may change their position and/or orientation relative to the magnetometer, resulting in a net measured magnetic field change. In one embodiment, the magnetometer may be embedded in a magnetic material to form an integrated sensor. In an alternative embodiment, the magnetic material and the magnetometer are separate. The magnetic field data received by the magnetometer is analyzed to provide useful information for force and contact location. The classification algorithm for analyzing the output of the magnetometer can locate the pressure, accuracy>98 percent. In some embodiments, the regression algorithm may localize the pressure to an average of about 3mm²The area of (a). In this regard, systems and methods for sensing deformation of magnetic materials, such as sensing skin, can meet the increasing demand for sensors for fields such as robotic manipulation, soft systems, and wearable devices that are easy to manufacture, fast to integrate, and rich in information.

Drawings

FIGS. 1A-1E illustrate various embodiments of a sensor.

Fig. 1F is a schematic diagram of a data processing procedure.

FIGS. 2A-2F are a series of graphs depicting exemplary results of sensing.

Fig. 3A-3G are a series of graphs depicting alternative exemplary results of sensing.

Fig. 4A-4C are sensor examples according to several alternative embodiments.

FIG. 5 is a series of graphs showing the classification and regression results.

Fig. 6 shows X and Y vectors for an example implementation on a robotic arm.

Fig. 7 is a visual vector of magnetic material indentation.

FIG. 8 is a flow chart describing a manufacturing technique.

FIG. 9 is an alternative embodiment of the sensor, in which the magnetometer is separate from the magnetic material.

Fig. 10 shows another alternative embodiment.

Detailed Description

In one exemplary embodiment, the sensor 100 includes a magnetic material 101 and a magnetometer 102, the magnetometer 102 being capable of sensing changes in the magnetic field of the magnetic material 101 resulting from deformation of the material 100. In the exemplary embodiment shown in fig. 1A-1B, the sensor 100 is applied as tactile skin with a fixed retractable triaxial magnetometer 102 covered by a soft elastomer 103 embedded with a dispersion of magnetic particles 104, thereby forming a magnetic material 101. The composite magnetic material 101 maintains the scalability and flexibility of the body elastomer 103 and is compatible with the scalable circuitry. In an alternative embodiment, multiple magnetometers 102 may be used (see FIGS. 1C-1D). When a deformation is applied to the surface of the sensor 100, the magnetic particles 104 are displaced relative to the rest position of the magnetometer 102. (see FIG. 1E) the magnetometer 102 measures changes in the surrounding magnetic field and analyzes the data to determine the location and force of the contact. The magnetometer 102 measures its surrounding magnetic field in the x, y and z directions. The plurality of magnetic particles 104 distributed throughout the magnetic material 101 is considered as input data, which is ultimately reduced to three-axis magnetic field measurements to retain information about the deformation of the material 101. Morphological calculations may also be employed by the inherent dimensionality reduction of material 101 itself. For example, the sensor 100 may employ morphological computational properties to inherently reduce the dimensionality of the output prior to analysis, thus eliminating the need for dense arrays of underlying microelectronic chips and wiring.

The overall magnetic strength of the magnetic material 101 is less compared to discrete permanent magnets or other conventional magnets. The signal amplitude is still sufficient to locate the touch location and estimate the force on the surface of the sensor 100. Furthermore, incorporating the particles 104 into the elastomer 103 allows the sensor 100 to have few limitations in shape, size, or thickness. Since it does not require a multilayer molding process as in some magnetic sensors, the manufacturing process thereof is also simplified.

Referring again to FIGS. 1C-1D, there is shown a sensor 100 having a plurality of magnetometers 102, with one magnetometer 102 identified as a reference magnetometer 105. In the embodiment shown in fig. 1C-1D, five magnetometers 102 are located adjacent to the magnetic material 101, while the reference magnetometer 105 is located at a distance from the magnetic material 101. In this particular example, the five magnetometers 102 are 15mm apart, which is the range before the closest magnetometer 102 can no longer detect the signal of the magnetic material 101. Each 15mm range overlaps the range of another magnetometer by 2.5mm to maximize the functional surface area and minimize the number of magnetometers 102 required. The reference magnetometer 105 is used as a reference for measuring a magnetic signal (i.e., ambient magnetic field noise) different from the magnetic material 101. With multiple magnetometers 102 and a single reference magnetometer 105, the sensor 100 is able to filter ambient magnetic field noise and motion and is able to incorporate output data analysis to address the non-linear increase in the system. In other words, the signal of the reference magnetometer 105 combined with the main magnetometer 102 isolates the change in magnetic flux due to the deformation of the magnetic material 101. The magnetic flux signal is evaluated to provide a real-time estimate of force and position.

In the exemplary embodiment shown in fig. 1C-1D, the signals may be evaluated by a trained neural network to provide an estimate of the force and location of the contact on the sensor 100. FIG. 1F is a schematic diagram of the preprocessing steps for contact location and force estimation. In these pre-processing steps, the raw magnetometer values are individually calibrated, transformed from the reference signal, filtered, and scaled to the neural network input.

In more detail, in one embodiment, the signal processing combines calibration and pre-processing to minimize the amount of data collection necessary while keeping the neural network inputs limited to the raw magnetometer data. Each magnetometer 102 outputs triaxial data about its surrounding magnetic field. For the embodiment illustrated by fig. 1-2, the six

magnetometers

102, 105 appear as a total of 18 data points for each sample. For pre-calibrated magnetometers 102, 105 (which provide offset and scaling), these parameters can be applied to the raw data to calibrate the signals separately. The offset may be determined by the average between the maximum and minimum signals in each direction. The scaling may be determined by dividing the average arc length of all three directions by the average arc length of each direction. Next, the affine transformation of the reference magnetometer 105 is applied to the five magnetometers 102. This transformation allows motion and environmental noise to be removed due to position and environmental noise if the reference magnetometer 105 is kept fixed relative to the other magnetometers 102. Noise removal allows data acquisition on one plane. After calibration and filtering of the data, the data is prepared for neural network input by removing the mean and scaling to the unit variance determined by the training data. While this system provides useful information, it is susceptible to the additional noise present in a multi-magnetometer 102, 105 system. Thus, a multi-tier perceptron implemented with mlpripressor in skleran may be used.

The magnetic particles 104, which may include magnetic Ne-Fe-B particles or nanoparticles, are approximately 200 μm or less in diameter as compared to conventional techniques using large rigid magnets. Nevertheless, particles 104 of different sizes and shapes may be used, depending on the intended application, as long as the composite magnetic material 101 retains a certain degree of scalability or flexibility. Depending on the intended application and the number of magnetic particles 104 used, the composite magnetic material 101 may have the same or similar properties as the elastomer 103 used to form the composite material 101. In addition, the use of micro-scale magnetic particles 104 may reduce the strength of internal stress concentrations when mechanical loads are applied to the magnetic material 101, and also provide flexibility and/or scalability to the material 101. For example, when a large magnet is embedded in an elastomer, delamination may occur at the interface between the hard and soft elastomers under mechanical load due to the difference in compliance between the two materials. Furthermore, such embodiments can enable thin geometries and/or contain sharp 3D geometries.

In one embodiment, the sensor 100 is formed in the following manufacturing process (see FIG. 8). First, the silicone elastomer 103 is mixed with the magnetic fine particles 104, and the composite material is cured under a magnetic field, thereby functionalizing the magnetic material 101. The material is cured in a magnetic field to align the magnetic particles 104 prior to embedding into the cured elastomer 103, resulting in a uniform magnetic orientation of the magnetic particles 104. In an alternative embodiment, a non-uniform magnetic orientation is used. As a further example, the prepolymer and crosslinker can be shear mixed at a ratio of 1:1 for about 30 seconds. The pre-cured elastomer mixture may be manually mixed with the magnetic particles 104 (MQP-15-12; Magnequench) in a 1:1 weight ratio to form the magnetic material 101. Then, the uncured magnetic material 101 was poured into a mold and degassed for 5 minutes. A thin plastic film may be placed on top of the mold and excess material 101 may be expelled. The mold may then be filled and then inverted onto the surface of a permanent magnet (N48; Applied Magnets). The material 101 may then be cured at room temperature and removed from the mold within an hour. Finally, the magnetic material 101 may be adhered (Silpoxy; smoothen-On) On top of a plate (e.g. MLX 90393; Sparkfun) of a commercial magnetometer 102. In an alternative embodiment, polyurethane foam is used as elastomer 103; however, those skilled in the art will understand that: several types of elastomers 103 may be used. In yet another alternative embodiment, a deformable material other than a polymer may be used as the substrate for the composite magnetic material 101.

With respect to the design and manufacture of the scalable circuit 106, the scalable circuit 106 may include a magnetometer 102(MLX 90393; Melexis) and five output lines (FIG. 1B). For the second 3.3 volt line, an additional output line may be used, which may be useful due to the single layer design of the scalable circuitry 106. A thin layer of copper and chromium was sputtered onto the PDMS (Sylgard 184; Dow Corning) surface and laser-mapped to leave traces of the circuitry. Eutectic indium gallium (EGaIn) selectively wets the remaining copper traces when immersed in the NAOH solution. The circuit components 106 may then be placed directly over the liquid metal traces and sealed with an additional PDMS layer.

For example, a sensor 100 constructed in this manner can collect data, such as pressure data, resulting from deformation of the magnetic material 101 and changes in the magnetic field from the material 101, which data is sensed by the magnetometer 102. In some embodiments, due to the non-uniform distribution of particles 104 within the magnetic material 101 that may generate unique magnetic fields, data-driven techniques may be used to address the locations of deformations that occur on the magnetic material 101Classification is performed and the depth of this deformation is estimated. In particular, classifying the localization has an accuracy of 98%, for example, for a 5 × 5 grid of 3mm resolution and a 5mm radial circle with 3 discrete depths. The regression algorithm may localize the contact to 3mm²The area of (a). In this regard, some embodiments disclosed herein provide a method that addresses the need for a continuous and soft tactile surface that is easy to manufacture, quickly integratable, and adaptable to geometry.

As an example of position sensing, for embodiments containing a 5 x 5 grid, force and magnetic field changes are collected on a 5 x 5 grid at 3mm resolution up to 3mm depth (see fig. 2A), for a total of 25 classes. Using a uniform random distribution, 2750 contact samples were collected at these 25 locations. Each category (25 categories total) includes approximately 100 samples.

Several different classification algorithms are capable of accurately distinguishing 25 locations, as shown below. In this regard, classification results using Quadratic Discriminant Analysis (QDA) are discussed herein to illustrate the performance of various aspects. In the case of a misclassification, the predicted class is adjacent to the true position (see fig. 2B, QDA classification for position 13). Fig. 2C shows the classification accuracy for each location (all QDA classification results are grouped by category). Fig. 2D-2E show the mean absolute error from linear regression grouping for x-position, y-position, and the mean absolute error from KNN regression of forces, respectively.

To estimate the position, the 25 discrete positions are converted into their coordinate positions. For the 5 × 5 grid and linear regression experiments, the average error for the x position was 1.1mm and the average error for the y position was 3.8 mm. The output estimate near the sensor edge may have a lower accuracy and a higher standard deviation. Since the magnetic signal is 1/d3 with respect to distance, the signal quality is expected to decrease with increasing distance. At these points along the edge, the random distribution of particles may start to have a greater effect on the output signal than the applied deformation. This may lead to abnormal signal changes and may also be a reason why data-driven techniques are more useful than function fitting methods in certain non-limiting embodiments.

As an example of position and depth sensing, for embodiments comprising a circular sensor 100, force-controlled changes in the magnetic field were measured for 8 different XY positions and 3 different depths (dZ ═ 1, 2, or 3mm) (see fig. 3A). Using a uniform random distribution, 2850 contact samples were collected for these 24 XYZ locations. There are approximately 110 samples per category (24 categories total). As shown, Quadratic Discriminant Analysis (QDA) can be used to classify locations based on XY location and depth. If the prediction class is wrong, it can be predicted as a neighboring class (see FIG. 3B for QDA classification results at position 3 and 1mm depth). Misclassifications between adjacent positions may be more common than misclassifications between adjacent depths. Greater correlation between z-axis magnetic field and pressure can be used to distinguish depth. Since all test sites are closer to the magnetometer 102 than in the 5 x 5 experiment, the same noise introduced from the particle 104 may not be perceived or may not be present. The classification accuracy for each location is shown in fig. 3C. Generally, the smaller the applied pressure (depth 1), the smaller the signal variation and the lower the accuracy. For this sample, the classification of position 3 and depth 1 has a lower accuracy. This may be due to the combination of small offsets that result in the right signal, which is also evident in fig. 3D (average absolute error of linear regression output grouped by position for x position) and fig. 3E (y position) for large errors at

positions

2, 3 and 4. Fig. 3F shows the mean absolute error for the z position, and fig. 3G shows the mean absolute error for the KNN regression grouped by force position.

With continued reference to the above example, the 24 classes are converted to their true coordinates (x, y, z) for position estimation. For 8-point circle and linear regression, the mean absolute error for the x position was 1.2mm and the mean absolute error for the y position was 3.4mm in all categories. The difference in error between the x and y coordinates may imply a slight misalignment in this test, as evidenced by the change in position error in fig. 3D and 3E. The z position error is relatively small (0.03mm), probably due to the large signal variation associated with the 1mm depth variation (fig. 3F).

To estimate the force, time series data and k-nearest neighbor (KNN) regression can be used. The inputs may be the Bx, By, and Bz components of the magnetic field, the internal temperature Bt of the magnetometer, and the load cell output at each time step. For the 5 x 5 grid demonstration, the average error of the force estimate is 0.44N (fig. 2F). For an 8-point circle, the average error of the force estimate is about 0.25N (FIG. 3G). The z-axis of the magnetic field is most strongly dependent on the applied pressure, making the force estimation relatively accurate. However, good signal variation may depend on the amount of deformation. Therefore, if the elastic body 103 for the magnetic material 101 has a higher young's modulus, the resolution of the force may be greater. The force applied during both trials ranged between approximately 0 and 2.5N, limited by the maximum depth of 3mm selected.

As an example of the capabilities of sensor 100, fig. 4A shows a simple 4-key directional game board. As shown in fig. 4A, four acrylic arrows are affixed to the surface of sensor 100 to assist the user in locating where pressure is applied to input directional commands. These four commands can be identified by the X, Y and Z component changes in the magnetic field. In this example no classifier is used, instead a simple threshold setting is sufficient when the button spacing is large enough. The positive and negative changes of X and Y are mapped to four arrow keys on the keyboard to play a game of "miss-to-bean (ms. pachman)" in the web browser. Example data from each direction of the game is also shown in FIG. 4A.

To demonstrate the speed and accuracy of the 5 x 5 mesh classifier, a mini game "mine sweeper" was played using a robotic controlled cylindrical ram, as shown in fig. 4B. Each of the 25 grid positions is mapped to a mouse position on the screen. The length of the signal (i.e., the duration of the applied pressure) indicates whether the user wants to click the left button to display the box or the right button to place a logo. Immediately after the signal returns to the stationary state, the position is predicted using the QDA classifier, and then the appropriate operation is performed. The raw data and classification results are shown in fig. 4B.

Since the sensor 100 is both scalable and flexible, it can be integrated with existing scalable circuit technology. Similar to the 4-key keyboard, four keyboard commands (ctrl + left, ctrl + right, ctrl + up, ctrl + down) are mapped to 4 positions (last, next, volume up, volume down) by vector thresholds to browse the music playlist (fig. 4C). Without the use of acrylic arrows, the location of the user input may vary, resulting in more noisy data. Furthermore, the user's hand and skin may deform along with the magnetic skin. Although both factors may generate additional noise, the system may still function using about 4 fundamental thresholds to determine the touch quadrant.

In other embodiments, the range and resolution of force and contact position may be increased by adjusting the manufacturing process of the sensor 100 or the magnetic material 101, modifying the training procedure, or adding additional magnetometers 102. The sensor 100 discussed herein can be used in applications including soft robotics, medical devices, steering, and tactile surfaces. Further, in certain non-limiting embodiments, the sensor 100 may be molded into a shape that conforms to the geometry of the host system and may be magnetically programmed to respond to a specified mechanical load or deformation.

With respect to some embodiments discussed herein, time series data is represented as a set of representative features. Furthermore, 21 features are identified manually instead of the automatic feature selection method. These 21 features include the minimum, maximum, mean, standard deviation, median and sum of each axis on the sample (18 features), and the scalar ratio between the three axes (3 features). At the end of the contact, features are calculated from the data collected during the contact, and classification and regression results are immediately output. Thus, as discussed herein, deformation of randomly distributed magnetic particles 104 may produce a repeatable and separable signal.

In analyzing the data received from the magnetometer 102, the magnetic field strength is predicted to decay with distance from the magnetometer 102 using the inverse cubic relationship:

with respect to the method, classification algorithms in the machine learning toolkit of scimit-left based on Python language were evaluated (see fig. 5 for a complete comparison of all available classification algorithms and implementation details). The results include reduced parameter scaling while enabling successful discrimination of multiple classes using a relatively small data set. By using the supervised learning algorithm, the supervised learning algorithm does not need a large amount of hyper-parameter settings and is very suitable for multi-class classification.

In this respect, the following classification algorithm is used:

LDA: linear discriminant analysis is a classifier that finds linear decision boundaries under the assumption that each class is a multivariate gaussian density with mean and same covariance. Singular Value Decomposition (SVD) is used without shrinkage, a priori or dimensionality reduction.

And (3) QDA: similarly, quadratic discriminant analysis is a classifier that finds quadratic decision boundaries between each class. Each class is modeled as a gaussian density and the output prediction is the class that is maximized by the bayesian rule. One key difference from LDA is: QDA does not assume that every class has the same covariance matrix.

KNN: the K-nearest neighbor classification algorithm classifies a new input using the K nearest samples separated by some distance metric. This is a common method of clustering data. Uniform weighting, manhattan distance (l1 norm), and k 5 were used in some of the experiments discussed herein.

RF: the random forest classifier fits a decision tree to the subsamples of the dataset. By randomly segmenting the data set, the classifier selects the best features from this subset.

DT: a Decision Tree (DT) classifier creates a binary tree and partitions the nodes based on the features containing the most information. In aspects discussed herein, a classification regression tree (CART) algorithm is used to use the decision tree.

GB: gradient Boosting (GB) is an integrated classifier that fits n regression trees to the gradient of a given loss function. In certain aspects discussed herein, 100 estimators, bias loss, and a learning rate of 0.1 are used.

As for the regression algorithm, the following regression algorithm was selected to estimate the XY coordinate position. With reference to some aspects discussed above, some regression algorithms are trained using the same features and samples as the previous classification (see fig. 6).

LR: linear regression fits a line using features as coefficients, while minimizing the sum of the remaining squares.

SVR: support vector regression fits a kernel function by minimizing the specified soft interval ε. Any error within this epsilon interval is considered zero and the L1 loss is calculated starting from this interval. In some cases discussed herein, epsilon-0.1 and rbf nuclei are used.

DTR: the decision tree regressor follows the same principle as decision tree classification by constructing a decision tree and partitioning nodes based on maximized information. However, in this regressor, the output is continuous.

KNN regressor: the KNN regressor expands the KNN classification scheme to a continuous output by using a weighted average of a continuous distance function.

With respect to the demonstration of the 5 x 5 grid discussed herein, several algorithms are able to distinguish 25 classes with 98% accuracy. It is worth noting that QDA requires more samples to achieve this accuracy than other algorithms. This may be due to the first 1000 samples not capturing the difference in features. The accuracy of LDA classification decreases as the sample size increases. This may mean that the features may not be linearly separated, as more samples may increase noise. The space for these features can be increased making them easier to separate linearly.

For the 8-point circle demonstration, there are several algorithms that can distinguish between 24 categories. Generally, decision tree based approaches work well. Similar to the 5 × 5 trellis results, QDA requires more samples before achieving similar performance as the CART, RF, and GBoost methods. KNN also performs well in these cases. This may be due to the z-direction magnetic field being classified by magnitude into categories [0,7], [8,15] and [16,23 ]. This allows the cluster to quickly reduce the problem to 8 options. Because the radius of the circle is less than the 5 x 5 grid, the same noise is not seen in the material particle distribution discussed previously.

The linear regression algorithms (linear, ridge, lasso, elastic) all have a continuous output that can estimate X positions with an average error of about 1.1mm and Y positions with an average error of about 2.5 mm. The KNN and DT results may be affected by the quasi-discrete nature of the data input, resulting in a quasi-discrete output. However, continuous sampling may be used over the entire surface.

With respect to the raw vectors of the 5 x 5 grid, the magnetometer 102 may have an internal coordinate system, which may be determined by the size and direction of the output vectors. For example, the x-axis of the magnetometer 102 crosses in FIG. 6, and the output may change from negative to positive. Each quadrant around the signal may reflect the correct sign based on its position, and the signal amplitude decreases with distance from the magnetometer 102. However, some differences from this pattern may occur at the edges of the grid experiment.

In FIG. 7, the maximum Bx and Bt vectors are plotted at 10mm/min for each position of the grid pattern of 3mm indentations at 25 positions. In fig. 7, differences are marked with asterisks. It should be noted that the inner 9 position signals are large relative to the edge positions, and the full vector is not shown in the box. It is clear, however, that the inner 9 positions follow the expected sign of the x-axis and y-axis of the magnetometer 102. Specifically, the X vector follows a (positive, negative) pattern from left to right when traversing the y-axis. Similarly, the Y vector follows a (negative, positive) pattern from top to bottom when crossing the x-axis. These patterns are expected and follow an approximate theory.

The edge situation is slightly different. For example, consider the X vectors (red) in

locations

15 and 20. Although they are located on the negative side of the x-axis (shown by the center 9 positions), the signals are positive. This difference only occurs at the edge positions. The deformation imposed at this distance from the magnetometer 102 may be outweighed by the sample inconsistency. Due to the inverse cubic relationship, the signal decays rapidly with increasing distance. In other words, the displacement of the polymeric particles under the indenter has a greater effect on the net magnetic field change than the overall displacement. A similar effect can be seen in the Y vector (green) at

positions

20 and 24 where the signal should be positive but in practice relatively small and negative. These relatively unpredictable differences may motivate the use of data-driven techniques, as compared to model-based techniques.

In other non-limiting embodiments, magnetometer 102 can be used to sense the deformation of composite magnetic material 101 as it moves relative to composite magnetic material 101 during operation. For example, in at least one non-limiting embodiment, the composite magnetic material 101 can be located on a gripper or robot arm of a robotic arm while the magnetometer 102 is located on another portion of the robotic arm, such as an elbow, shoulder, base, or other location. During operation of such exemplary embodiments, the gripper or robot arm may move simultaneously with other aspects of the robotic arm (including the aspect that includes magnetometer 102), thereby moving composite magnetic material 101 and magnetometer 102 relative to each other. In this regard, the magnetometer 102 may be used to sense the deformation of the composite magnetic material 101 as it moves around during operation. In other embodiments, the magnetometer 102 can be located in a different location than the robotic arm or other device attached to the composite magnetic material, such as on an object to be manipulated by the robotic arm.

In the embodiment shown in fig. 9, the magnetic material 101 is attached to a key, where the magnetometer 102 is part of the robot gripper. The magnetometer 102 in the robot holder can be positioned with a millimeter-sum (sum-mm) accuracy to the magnetic material 101 on the key and enables the robot to pick up objects in the same way at the same location each time. Furthermore, the robot can consistently position repeatable grip and object poses even before contact. Magnetometer 104 can be implemented in small size (7 x 2mm), provides fast sampling frequency (>100Hz), and can be easily integrated into a system through serial communication.

Fig. 10 depicts an alternative embodiment of a sensor 100, wherein the magnetometer 102 is separated from the magnetic material 101, allowing for 3D localization. By separating the magnetic material 101 from the magnetometer 102, the robot is free to move and measure ambient magnetic flux changes due to motion and deformation. The sensor 100 may supplement the positioning based on visual objects. In one exemplary embodiment of sensor 100 with separate components, magnetometer triad 102 is mounted on a circuit board having four input lines for SDA, SCL, 3.3V, and GND. These four lines allow the magnetometer 102 to use i²c and microcontrol attached to end effectors or grippersThe devices communicate.

For the sensor 100 depicted in fig. 9-10, positioning and force feedback are controlled by maxwell's electromagnetic equations. For some applications, the process may be simplified by estimating the shape of the magnetic field above the magnetic material 101 as a 2D gaussian distribution. With this basis, the z-component of the magnetic field at the surface of the magnetic material 101 can be determined and measured by the magnetometer 102. For example, for a thickness of 2mm, the magnetic material 101 used in the present exemplary embodiment ranges between 3500 and 4500 μ T and serves as a reasonable boundary for a least squares fit.

As the robot gripper and magnetometer 102 pass near the magnetic material 101, the position of the gripper is recorded when a maximum magnetic field is encountered. Moving the gripper to this position will center the magnetic material 101 on the axis. By repeating this process in the other direction, the robot gripper can be centered on the magnetic material 101.

Alternatively, a complementary vision-based system would be able to position the robot gripper near the object to which the magnetic material 101 is attached, but without information about the scan direction. In this case, a short range scan can be performed in any direction and a 1D gaussian distribution is fitted to the data points by non-linear least squares optimization. This process may then be repeated for the second axis. By performing these steps, the robot gripper can be positioned at the peak of the gaussian distribution estimate. Once the mechanical gripper containing magnetometer 102 is positioned to the central axis, maxwell's equations can be used to estimate the position of the surface of magnetic material 101. The robot gripper can then be moved incrementally to access the magnetic material 101 surface.

In additional embodiments, the system may include multiple magnetometers at any number of locations to sense one or more deformations in composite magnetic material 101. In yet another embodiment, the plurality of magnetic particles 104 may include materials such as neon (Ne), iron (Fe), boron (B), neodymium (Nd), samarium (Sm), cobalt (Co), and any suitable combination thereof. Further, in some non-limiting embodiments, the plurality of magnetic particles comprises microparticles, including particles having a size of 10^-7.5Rice to 10^-4.5Particles in the rice rangeAnd particles ranging in size from 0.5 μm (micrometers) to 0.5mm (millimeters), and/or nanoparticles, including particles having a size less than 700nm (nanometers).

The features disclosed in the foregoing description, or the following claims, or the accompanying drawings, expressed in their specific forms or in terms of a means for performing the disclosed function, or a method or process for attaining the disclosed result, as appropriate, may, separately, or in any combination of such features, be utilised for realising the invention in diverse forms thereof. In particular, one or more features of any embodiment described herein may be combined with one or more features of any other embodiment described herein.

Protection may also be sought for any feature disclosed in any one or more of the publications referenced and/or incorporated by reference in connection with the present disclosure.

Claims

1. A sensor, the sensor comprising:

a magnetometer; and

a composite magnetic material having a magnetic field, the composite magnetic material comprising:

a deformable material; and

a plurality of magnetic particles dispersed within the deformable material;

wherein the magnetometer is configured to sense the magnetic field of the composite magnetic material.

2. The sensor of claim 1, wherein the magnetometer is configured to sense changes in the magnetic field of the composite magnetic material caused by deformation of the composite magnetic material.

3. The sensor of claim 1, wherein the deformable material is an elastomer.

4. The sensor of claim 1 or 2, further comprising:

a reference magnetometer positioned remote from the magnetic field of the magnetic material and configured to sense a background magnetic field.

5. The sensor of claim 1, wherein the magnetometer is in a fixed position relative to the composite magnetic material.

6. The sensor of claim 1, wherein the magnetometer is configured to remain in a substantially fixed position relative to the composite magnetic material in response to deformation of the composite magnetic material.

7. The sensor of claim 1 or 2, wherein the composite magnetic material is in contact with the magnetometer.

8. The sensor of claim 1, wherein the magnetometer is a three axis magnetometer.

9. The sensor of claim 1 or 2, wherein the plurality of magnetic particles are 200 μ ι η or less.

10. The sensor of claim 1 or 2, wherein the plurality of magnetic particles have substantially the same kind of magnetic orientation.

11. The sensor of claim 1 or 2, wherein the plurality of magnetic particles have substantially different kinds of magnetic orientations.

12. The sensor of claim 1, wherein the plurality of magnetic particles are substantially non-uniformly distributed throughout the deformable material.

13. The sensor of claim 1 or 2, wherein the composite magnetic material is substantially stretchable.

14. A sensor according to claim 1 or 2, wherein the composite magnetic material retains the material properties of the deformable material.

15. The sensor of claim 1 or 2, further comprising:

a scalable magnetometer circuit connected with the magnetometer.

16. The sensor of claim 15, wherein the scalable magnetometer circuit is part of the composite magnetic material.

17. A sensor according to claim 1 or 2, wherein a change in the magnetic field of the composite magnetic material occurs in response to the deformation, wherein the deformation causes at least one of the plurality of magnetic particles to change its position relative to the magnetometer.

18. A method of positioning an object comprising soft magnetic material, the method comprising:

passing a magnetometer within the measurable magnetic field range of the magnetic material;

determining a maximum magnetic field measurement to locate a central axis of the magnetic field;

aligning the magnetometer with the central axis;

repeating the above steps to find a minor axis; and

and determining the distance between the surface of the object and the magnetometer according to the difference value between the magnetic field intensity obtained by measuring the magnetic field intensity on the surface of the magnetic material and the magnetic field intensity obtained by calculating.

19. A method of determining a position or contact force using a soft touch sensor, the method comprising:

receiving a signal from a magnetometer, wherein the signal comprises a measurement of a magnetic field emitted by a magnetic material comprising an elastomer and a plurality of magnetic particles dispersed within the elastomer;

calibrating the signal;

converting the signal;

filtering the signal;

scaling the signal for input into a neural network; and

the position or contact force is obtained using a neural network.

20. The method of claim 19, wherein filtering the signal comprises:

background signals from a reference magnetometer are used to remove motion noise and environmental noise from the signals.